Safety4DM vision-language framework for risk-aware dynamic maps in collaborative driving
Original title: Safety4DM: A Vision-Language Framework for Risk-Aware Dynamic Maps in Vehicle-Road-Cloud Collaborative Driving
Fine-tuned Qwen3-VL-4B reaches 89.83% accuracy on RoadRisk dataset with SafeStep interface achieving over 93% query accuracy.
The paper introduces Safety4DM, a safety-oriented language system for vehicle-road-cloud dynamic maps. It includes RoadRisk dataset and SafeStep prompting mechanism for driving advice. Experiments on SafeVRC-Sim and SafeDM-QA benchmarks show strong performance across multiple LLM families.
- RoadRisk dataset covers four road conditions and four risk levels
- SafeStep achieves over 93% accuracy with models 20B and above
- Compact Qwen3-VL-4B reaches 89.83% after fine-tuning
- SafeVRC-Sim simulation framework developed
From the original article
1 min read at arxiv.orgDynamic map (DM) is a core information infrastructure for vehicle-road-cloud (VRC) cooperative autonomous driving. It fuses perception data from multiple sources to overcome the inherent line-of-sight and field-of-view limits of single-vehicle perception. However, DM still falls short on driving safety in two ways.
By Zhengshu Zhou · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.